Article Preview
TopIntroduction
In recent years, number and character recognition has been gaining widespread attention in the field of image processing, thanks to the rapid development of digital technology. The segmentation of numbers and characters in dial images is a popular research directions in this field. Dials are widely used in instruments, automotive dashboards, and other fields, making it essential to accurately segment numbers and characters for improving recognition accuracy and practical applications. This study aims to achieve precise segmentation of characters and numbers in dial images using methods such as the Hough transform and binary wavelets, providing effective technical support for related applications (Fu et al., 2001).
There have been noteworthy research results in the field of digit and character segmentation, we present an overview of noteworthy research results in the field of digit and character segmentation from 2021 to 2023, covering various methods and techniques. The method proposed by Li and Dong (2022) utilizes a segmentation approach based on nearest neighbor values and prior knowledge of license plate characters to obtain the first five complete characters of a license plate. The remaining five characters of the license plate are obtained through character concatenation based on an enumeration method, achieving a complete license plate segmentation.
Wang and Li’s (2023) method first used ROI regions and then connected component processing algorithms on images of exam paper score columns, which are acquired using a high-resolution scanner. The abbreviation “ROI” stands for “Region of Interest,” referring to specific areas in an image or graphic that are of particular interest or focus for processing. This process extracts handwritten scores, forming individual images, which are then normalized and formatted into a CSV file. These data are fed into a pre-trained LeNet-5 network for recognition. During the training of the LeNet-5 network, parameters such as batch size, learning rate, epochs, and weight quantities are continuously adjusted, resulting in a recognition accuracy of 93.2% (Wang & Li, 2023).
In this study, we first use the Hough transform to perform skew correction on dial images with fixed formats. Subsequently, we employ binary wavelets to “scale” the images, achieving coarse segmentation of characters or numbers. Additionally, we propose a binary threshold iteration method to accurately determine the position of each character or number even when it undergoes adhesion or fragmentation, which results in precise segmentation. The results show that the proposed method achieves a recognition rate of 98.5% for both letters and numbers in 98 fixed-format dial images, validating its efficiency and accuracy. It can provide strong support for the practical application of character and number segmentation technology.